An Empirical Study of Depression,Anxiety and Stress using MachineLearning

R. Vandana¹, K.Balaraju², N. Drakshayani³, P.V. Bhanu Prakash⁴, G. Narendra Kumar5
Supervisor: Dr. G. SITARATNAM, Professor &HOD, Dept. of CSE-DS, VIET
¹Department of CSE (DS), Visakha Institute of Engineering and Technology, Andhra Pradesh, India
2Department of CSE (DS), Visakha Institute of Engineering and Technology, Andhra Pradesh, India
3Department of CSE (DS), Visakha Institute of Engineering and Technology, Andhra Pradesh, India
4Department of CSE (DS), Visakha Institute of Engineering and Technology, Andhra Pradesh, India
5Department of CSE (DS), Visakha Institute of Engineering and Technology, Andhra Pradesh, India

Abstract - This study develops a machine learning–assisted analytical system for evaluating psychological distress using the 42-item Depression Anxiety Stress Scales (DASS-42). The proposed system implements an end-to-end pipeline encompassing data preprocessing, standardized scoring, visualization, and multi-class classification of depression severity levels. A Random Forest classifier (Breiman , 2001) was employed to predict depression severity directly from raw questionnaire responses.The model was trained and evaluated using a stratified train-test split on a DASS-42 dataset. It achieved an overall accuracy f 86.67% on the held-out test set. Performance was excellent for the Normal (100%) and Extremely Severe (93.5%) categories, and acceptable for the Severe category (75.0%). However, classification of Mild and Moderate levels remained challenging, primarily due to class imbalance and limited sample representat ion within theseintermediate categories.The findings demonstrate the feasibility of integrating machine learning with validated psychometric instruments such as the DASS-42 (Lovibond & Lovibond, 1995) to enable scalable and automated mental health screening. Feature importance analysis further revealed key questionnaire items contributing to predictions. While the system shows promise for identifying extreme severity levels, the study underscores the need for balanced datasets and advanced techniques (e.g., resampling or cost-sensitive learning) to improve performance across all severity categories.Key Words: Depression Anxiety Stress Scales (DASS-42); machine learning; Random Forest; mental healthassessment; depression severity classification; class imbalance